Set Reference Multi Latent - ComfyUI Built-in Node Documentation

This node encodes one or more reference images and appends their latents to the conditioning that an edit model consumes. It is the multi-image counterpart of Set Reference Latent: you connect a VAE plus a set of reference images, and the node attaches every encoded latent to the positive conditioning and, when one is connected, to the negative conditioning as well. Use it when a model expects several reference images at once instead of a single latent.

Inputs

Common Inputs

Parameter	Description	Data Type	Required	Range
`positive`	The positive conditioning to modify with the reference latents.	CONDITIONING	Yes	-
`negative`	Optional negative conditioning that receives the same reference latents. Leave it unconnected to reuse the modified positive conditioning on the `negative` output.	CONDITIONING	No	-
`vae`	The VAE used to encode the connected reference images. When no VAE is connected the reference images are ignored.	VAE	No	-
`resolution`	How each reference image is prepared before encoding. `"keep resolution"` encodes it at its incoming size; `"resize"` scales it to a target megapixel count while preserving the aspect ratio, which reveals `megapixels`. (default: “keep resolution”)	DYNAMIC_COMBO	Yes	`"keep resolution"` `"resize"`
`images`	Growable slot: connect one or more reference images (`image_1`, `image_2`, …) to use as references. Images are encoded in slot order and appended to the conditioning in that order.	IMAGE	No	1 to 100

Resize Inputs

These options appear when resolution is set to "resize".

Parameter	Description	Data Type	Required	Range
`megapixels`	Target size of each reference image, preserving aspect ratio. 1.0 is about 1024 x 1024 pixels. (default: 1.0)	FLOAT	Yes	0.01 to 16.0

Outputs

Output Name	Description	Data Type
`positive`	The positive conditioning with the encoded reference latents appended.	CONDITIONING
`negative`	The negative conditioning with the same reference latents appended. Returns the positive output when no negative conditioning is connected.	CONDITIONING
`empty_latent`	The first reference latent multiplied by 0.0, ready to seed a sampler. When no reference latent is produced, a square empty latent at the selected resolution is returned instead.	LATENT

Notes

The latents are appended rather than replaced, so chaining several Set Reference Multi Latent nodes accumulates references.
Each connected image is encoded on its own, so the reference images do not have to share a size.
Without a connected VAE or without any connected image, positive and negative are returned unchanged and empty_latent is a square empty latent: 1024 x 1024, or sized from megapixels when resolution is "resize".

This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub

Source fingerprint (SHA-256): 06f98e7c21e8750a2bdf365cb80fac990bcf01246f183c60ef83bd195fc1225b

​Inputs

​Common Inputs

​Resize Inputs

​Outputs

​Notes

Inputs

Common Inputs

Resize Inputs

Outputs

Notes